You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: citing/index.html
+9-4Lines changed: 9 additions & 4 deletions
Original file line number
Diff line number
Diff line change
@@ -93,23 +93,28 @@
93
93
<divclass="section" itemprop="articleBody">
94
94
95
95
<!-- when new paper is published, add that paper and say "if you use SWRF, MultiSWRF/MultiSWRF*, or MultiSWRFDB/MultiSWRFDB*, cite the following paper" -->
96
-
<p>If you use <strong>scikit-rebate</strong> or the <strong>MultiSURF</strong> algorithm in a scientific publication, please consider citing the following paper (currently available as a pre-print in arXiv):</p>
96
+
<p>If you use <strong>scikit-rebate</strong> or the <strong>MultiSURF</strong> algorithm in a scientific publication, please consider citing the following paper:</p>
97
97
<!-- *Urbanowicz, Ryan J., Randal S. Olson, Peter Schmitt, Melissa Meeker, and Jason H. Moore. "Benchmarking relief-based feature selection methods." arXiv preprint arXiv:1711.08477 (2017).* -->
98
98
<p><em>Urbanowicz, Ryan J., Randal S. Olson, Peter Schmitt, Melissa Meeker, and Jason H. Moore. "Benchmarking relief-based feature selection methods for bioinformatics data mining." Journal of Biomedical Informatics, 85:168–188, 2018.</em></p>
99
99
<p>Alternatively a complete <strong>review of Relief-based algorithms</strong> is available at:</p>
100
-
<p><em>Urbanowicz, Ryan J., Melissa Meeker, William LaCava, Randal S. Olson, and Jason H. Moore. "Relief-based feature selection: introduction and review." arXiv preprint arXiv:1711.08421 (2017).</em></p>
100
+
<!-- *Urbanowicz, Ryan J., Melissa Meeker, William LaCava, Randal S. Olson, and Jason H. Moore. "Relief-based feature selection: introduction and review." arXiv preprint arXiv:1711.08421 (2017).* -->
101
+
<p><em>Urbanowicz, Ryan J., Melissa Meeker, William LaCava, Randal S. Olson, and Jason H. Moore. "Relief-based feature selection: introduction and review." Journal of Biomedical
102
+
Informatics, 85:189–203, 2018.</em></p>
101
103
<p>To cite the <strong>original Relief</strong> paper:</p>
102
104
<p><em>Kira, Kenji, and Larry A. Rendell. "A practical approach to feature selection." In Machine Learning Proceedings 1992, pp. 249-256. 1992.</em></p>
103
105
<p>To cite the <strong>original ReliefF</strong> paper: </p>
104
106
<p><em>Kononenko, Igor. "Estimating attributes: analysis and extensions of RELIEF." In European conference on machine learning, pp. 171-182. Springer, Berlin, Heidelberg, 1994.</em></p>
105
107
<p>To cite the <strong>original SURF</strong> paper:</p>
106
-
<p><em>Greene, Casey S., Nadia M. Penrod, Jeff Kiralis, and Jason H. Moore. "Spatially uniform relieff (SURF) for computationally-efficient filtering of gene-gene interactions." BioData mining 2, no. 1 (2009): 5.</em></p>
108
+
<p><em>Greene, Casey S., Nadia M. Penrod, Jeff Kiralis, and Jason H. Moore. "Spatially uniform relieff (SURF) for computationally-efficient filtering of gene-gene interactions." BioData Mining 2, no. 1 (2009): 5.</em></p>
107
109
<p>To cite the <strong>original SURF*</strong> paper: </p>
108
110
<p><em>Greene, Casey S., Daniel S. Himmelstein, Jeff Kiralis, and Jason H. Moore. "The informative extremes: using both nearest and farthest individuals can improve relief algorithms in the domain of human genetics." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 182-193. Springer, Berlin, Heidelberg, 2010.</em></p>
109
111
<p>To cite the <strong>original MultiSURF*</strong> paper:</p>
110
112
<p><em>Granizo-Mackenzie, Delaney, and Jason H. Moore. "Multiple threshold spatially uniform relieff for the genetic analysis of complex human diseases." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 1-10. Springer, Berlin, Heidelberg, 2013.</em></p>
111
113
<!-- Add citations for SWRF* and μ-Relief below -->
112
-
114
+
<p>To cite the <strong>original SWRF*</strong> paper:</p>
115
+
<p><em>Stokes, Matthew E., and Shyam Visweswaran. "Application of a spatially-weighted relief algorithm for ranking genetic predictors of disease." BioData Mining, 5:20, 2012.</em></p>
116
+
<p>To cite the <strong>original μ-Relief</strong> paper:</p>
117
+
<p><em>Aggarwal, Nitisha, Unmesh Shukla, G. J. Saxena, et al. "Mean based relief: An improved feature selection method based on relieff." Applied Intelligence, 53:23004–23028, 2023.</em></p>
113
118
<p>To cite the <strong>original TuRF</strong> paper: </p>
114
119
<p><em>Moore, Jason H., and Bill C. White. "Tuning ReliefF for genome-wide genetic analysis." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 166-175. Springer, Berlin, Heidelberg, 2007.</em></p>
<p>We have designed the Relief-based algorithms to be integrated directly into scikit-learn machine learning workflows. Below, we provide code samples showing how the various Relief-based algorithms can be used as feature selection methods in scikit-learn pipelines.</p>
149
149
<p>For details on the algorithmic differences between the various Relief-based algorithms, please refer to <ahref="https://arxiv.org/abs/1711.08477">this research paper</a>.</p>
150
150
<h2id="using-the-core-algorithms">Using the Core Algorithms</h2>
151
-
<p>Core Relief-based algorithms are Relief-based algorithms that perform a single pass over the training data (i.e. each target instance is used one time). </p>
152
-
<p>ReliefF was the original, most widely-known core Relief-based algorithm and it allows you to specify the number of nearest neighbors to consider during feature scoring.</p>
153
-
<p>SURF, SURF*, MultiSURF, MultiSURF*, SWRF, SWRF*, MultiSWRF, MultiSWRF*, MultiSWRFDB, and MultiSWRFDB* are all extensions to the ReliefF algorithm that automatically determine the number of neighbors to consider when scoring the features. μ-Relief is an extension that, like ReliefF, requires a preset number of neighbors, but determines neighborhood membership differently than ReliefF.</p>
154
-
<p>The hyperparameter settings and usage examples for each of these algorithms are provided below. </p>
<p>Core Relief-based algorithms are Relief-based algorithms (RBAs) that perform a single pass over the training data (i.e. each target instance is used once). </p>
152
+
<p>ReliefF was the original, most widely-known core RBA and it allows you to specify the number of nearest neighbors to consider during feature scoring.</p>
153
+
<p>SURF, SURF*, MultiSURF, MultiSURF*, SWRF, SWRF*, MultiSWRF, MultiSWRF*, MultiSWRFDB, and MultiSWRFDB* are all core RBAs that automatically determine the number of neighbors to consider when scoring the features. μ-Relief is a core RBA that, like ReliefF, requires a preset number of neighbors, but determines neighborhood membership differently than ReliefF.</p>
154
+
<p>The hyperparameter settings and usage examples for each of these algorithms in scikit-rebate are provided below. </p>
155
+
<h3id="relieff">ReliefF</h3>
156
156
<!-- ReliefF is the most basic of the Relief-based feature selection algorithms, and the implementation allows you to specify the number of nearest neighbors to consider in the scoring algorithm. The parameters for the ReliefF algorithm are as follows: -->
157
157
<p>Determines neighborhood membership based on k-nearest neighbors (<code>n_neighbors</code>).</p>
158
158
<!-- Includes k nearest instances as neighbors (`n_neighbors`). -->
159
159
160
+
<p>To learn more about the algorithm, read this <ahref="https://doi.org/10.1007/3-540-57868-4_57">paper</a>.</p>
SURF, SURF\*, MultiSURF, MultiSURF\*, SWRF, SWRF\*, MultiSWRF, MultiSWRF\*, MultiSWRFDB, and MultiSWRFDB\* are all extensions to the ReliefF algorithm that automatically determine the ideal number of neighbors to consider when scoring the features. Note that all of these algorithms utilize the same group of hyperparameters, which are the same hyperparameters as ReliefF excluding `n_neighbors`. -->
<p>Includes instances closer than <code>μ-σ/2</code> as near neighbors.</p>
504
507
<p>Recomputes the mean distance and standard deviation per target instance (i.e. mean distance to the target instance and standard deviation of these distances).</p>
508
+
<p>To learn more about the algorithm, read this <ahref="https://doi.org/10.1016/j.jbi.2018.07.015">paper</a>.</p>
<p>Includes instances closer than <code>μ-σ/2</code> as near neighbors and instances farther than <code>μ+σ/2</code> as far neighbors. Instances within half a standard deviation of the mean distance are excluded from the neighborhood (are in the "deadband zone").</p>
615
619
<p>Recomputes the mean distance and standard deviation per target instance (i.e. mean distance to the target instance and standard deviation of these distances).</p>
620
+
<p>To learn more about the algorithm, read this <ahref="https://doi.org/10.1007/978-3-642-37189-9_1">paper</a>.</p>
<p>Adjusts weights given to neighbors through a sigmoidal gradient function. Weights decrease as distance to the target instance increases, and all instances are included in the neighborhood (instances farther than the global mean distance are added as far neighbors).</p>
837
842
<p>The global mean distance is the mean of all pairwise distances in the dataset and the global standard deviation, used in the gradient function, is the standard deviation of these distances.</p>
843
+
<p>To learn more about the algorithm, read this <ahref="https://doi.org/10.1186/1756-0381-5-20">paper</a>.</p>
<p>Includes as neighbors: the k (<code>n_neighbors</code>) instances whose absolute difference between 1) their distance from the target instance and 2) the mean distance among their class from the target instance is the greatest. </p>
1400
+
<h3id="-relief">μ-Relief</h3>
1401
+
<p>Includes as neighbors: the k (<code>n_neighbors</code>) instances whose absolute difference between 1) their distance from the target instance and 2) the mean distance among their class from the target instance is the greatest.</p>
1402
+
<p>To learn more about the algorithm, read this <ahref="https://doi.org/10.1007/s10489-023-04662-w">paper</a>. </p>
<p>2.) In very large feature spaces users can expect core Relief-based algorithm scores to become less reliable when run on their own. This is because as the feature space becomes very large, the determination of nearest neighbors becomes more random. As a result, in very large feature spaces (e.g. > 10,000 features), users should consider combining a core Relief-based algorithm with an iterative approach such as TuRF (implemented here), VLSRelief, or Iterative Relief. </p>
1875
1882
<p>3.) When scaling up to big data problems, keep in mind that the data aspect that slows down ReBATE methods the most is the number of training instances, since Relief-based algorithms scale linearly with the number of features, but quadratically with the number of training instances. This is the result of Relief-based methods needing to calculate a distance array (i.e. all pairwise distances between instances in the training dataset). If you have a very large number of training instances available, consider utilizing a class balanced random sampling of that dataset when running any ReBATE method to save on memory and computation time.</p>
1876
1883
<!-- References -->
1877
-
<!-- [^4]: *Urbanowicz, Ryan J., Randal S. Olson, Peter Schmitt, Melissa Meeker, and Jason H. Moore. "Benchmarking relief-based feature selection methods." arXiv preprint arXiv:1711.08477 (2017).* -->
1878
-
<divclass="footnote">
1879
-
<hr/>
1880
-
<ol>
1881
-
<liid="fn:1">
1882
-
<p><em>Kononenko, Igor. "Estimating attributes: analysis and extensions of RELIEF." In European conference on machine learning, pp. 171-182. Springer, Berlin, Heidelberg, 1994.</em> <aclass="footnote-backref" href="#fnref:1" title="Jump back to footnote 1 in the text">↩</a></p>
1883
-
</li>
1884
-
<liid="fn:2">
1885
-
<p><em>Greene, Casey S., Nadia M. Penrod, Jeff Kiralis, and Jason H. Moore. "Spatially uniform relieff (SURF) for computationally-efficient filtering of gene-gene interactions." BioData mining 2, no. 1 (2009): 5.</em> <aclass="footnote-backref" href="#fnref:2" title="Jump back to footnote 2 in the text">↩</a></p>
1886
-
</li>
1887
-
<liid="fn:3">
1888
-
<p><em>Greene, Casey S., Daniel S. Himmelstein, Jeff Kiralis, and Jason H. Moore. "The informative extremes: using both nearest and farthest individuals can improve relief algorithms in the domain of human genetics." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 182-193. Springer, Berlin, Heidelberg, 2010.</em> <aclass="footnote-backref" href="#fnref:3" title="Jump back to footnote 3 in the text">↩</a></p>
1889
-
</li>
1890
-
<liid="fn:4">
1891
-
<p><em>Urbanowicz, Ryan J., Randal S. Olson, Peter Schmitt, Melissa Meeker, and Jason H. Moore. "Benchmarking relief-based feature selection methods for bioinformatics data mining." Journal of Biomedical Informatics, 85:168–188, 2018.</em> <aclass="footnote-backref" href="#fnref:4" title="Jump back to footnote 4 in the text">↩</a></p>
1892
-
</li>
1893
-
<liid="fn:5">
1894
-
<p><em>Granizo-Mackenzie, Delaney, and Jason H. Moore. "Multiple threshold spatially uniform relieff for the genetic analysis of complex human diseases." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 1-10. Springer, Berlin, Heidelberg, 2013.</em> <aclass="footnote-backref" href="#fnref:5" title="Jump back to footnote 5 in the text">↩</a></p>
1895
-
</li>
1896
-
<liid="fn:6">
1897
-
<p><em>Stokes, Matthew E., and Shyam Visweswaran. "Application of a spatially-weighted relief algorithm for ranking genetic predictors of disease." BioData Mining, 5:20, 2012.</em> <aclass="footnote-backref" href="#fnref:6" title="Jump back to footnote 6 in the text">↩</a></p>
1898
-
</li>
1899
-
<liid="fn:7">
1900
-
<p>Aggarwal, Nitisha, Unmesh Shukla, G. J. Saxena, et al. Mean based relief: An improved feature selection method based on relieff. Applied Intelligence, 53:23004–23028, 2023. doi: 10.1007/s10489-023-04662-w. <aclass="footnote-backref" href="#fnref:7" title="Jump back to footnote 7 in the text">↩</a></p>
1901
-
</li>
1902
-
</ol>
1903
-
</div>
1884
+
<!-- [^1]: *Kononenko, Igor. "Estimating attributes: analysis and extensions of RELIEF." In European conference on machine learning, pp. 171-182. Springer, Berlin, Heidelberg, 1994.*
1885
+
[^2]: *Greene, Casey S., Nadia M. Penrod, Jeff Kiralis, and Jason H. Moore. "Spatially uniform relieff (SURF) for computationally-efficient filtering of gene-gene interactions." BioData mining 2, no. 1 (2009): 5.*
1886
+
[^3]: *Greene, Casey S., Daniel S. Himmelstein, Jeff Kiralis, and Jason H. Moore. "The informative extremes: using both nearest and farthest individuals can improve relief algorithms in the domain of human genetics." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 182-193. Springer, Berlin, Heidelberg, 2010.*
1887
+
[^4]: *Urbanowicz, Ryan J., Randal S. Olson, Peter Schmitt, Melissa Meeker, and Jason H. Moore. "Benchmarking relief-based feature selection methods for bioinformatics data mining." Journal of Biomedical Informatics, 85:168–188, 2018.*
1888
+
[^5]: *Granizo-Mackenzie, Delaney, and Jason H. Moore. "Multiple threshold spatially uniform relieff for the genetic analysis of complex human diseases." In European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, pp. 1-10. Springer, Berlin, Heidelberg, 2013.*
1889
+
[^6]: *Stokes, Matthew E., and Shyam Visweswaran. "Application of a spatially-weighted relief algorithm for ranking genetic predictors of disease." BioData Mining, 5:20, 2012.*
1890
+
[^7]: *Aggarwal, Nitisha, Unmesh Shukla, G. J. Saxena, et al. "Mean based relief: An improved feature selection method based on relieff." Applied Intelligence, 53:23004–23028, 2023. doi: 10.1007/s10489-023-04662-w.* -->
0 commit comments